Google makes LookML business definitions available to enterprise AI

Google has made LookML metadata indexing in Knowledge Catalog generally available, giving enterprise AI systems a route to the business definitions already maintained in Looker. Announced on 26 October, the feature applies to Looker Core on Google Cloud.[1]

The integration exposes more than database table names. LookML describes how measures are calculated, how data relates across tables and which filters define an analysis. Indexing that metadata makes those semantics available alongside catalogue information such as lineage and governance context for Gemini Enterprise and custom agents.[1]

Why knowing the table is not enough

A request for “revenue last quarter” can hide several decisions: whether revenue is gross or net, whether refunds are deducted, which date counts and which business units belong in the total. An AI model that sees only raw tables must infer those choices or ask someone to supply them. A maintained semantic layer can state them explicitly.[1]

Joins create another source of error. In a simple hypothetical, 100 orders each have three shipment records. Joining the tables without preserving the intended level of aggregation can count the same order value three times. A query may execute successfully and still report the wrong metric. The purpose of the semantic definition is to encode the relationships and calculation rules that prevent that mistake.

This is why the integration is relevant to agent systems. An agent can retrieve the organisation’s established definition before constructing an analysis or explaining a dashboard. It has a reference for what a metric means, rather than relying on the model’s general understanding of the word.[1]

The definition becomes a shared dependency

A semantic layer is valuable because an organisation can maintain a definition once and reuse it. That also makes a stale or ambiguous definition consequential: several dashboards and agents may inherit the same problem. Ownership, version history and a clear route for resolving competing definitions become part of the analytical workflow.

For an implementation, a useful test is a set of familiar business questions with accepted answers. The team can check which LookML definition the agent retrieved, whether it used the intended relationship and whether the final result agrees with the trusted calculation. This tests the connection between metadata access and actual behaviour.

Google’s release advances that connection by making established business semantics discoverable through the enterprise catalogue. It gives AI-assisted analysis a more concrete foundation: a definition the organisation has chosen and can inspect when the answer is disputed.[1]

Report an error or suggest a correction